Shifting material source of Chinese loess since ~2.7 Ma reflected by Sr isotopic composition

Deciphering the sources of eolian dust on the Chinese Loess Plateau (CLP) is fundamental to reconstruct paleo-wind patterns and paleo-environmental changes. Existing datasets show contradictory source evolutions of eolian dust on the CLP, both on orbital and tectonic timescales. Here, the silicate Sr and Nd isotopic compositions of a restricted grain size fraction (28–45 μm) were measured to trace the source evolution of the CLP since ~2.7 Ma. Our results revealed an unchanged source on orbital timescales but a gradual source shift from the Qilian Mountains to the Gobi Altay Mountains during the past 2.7 Ma. Both tectonic uplift and climate change may have played important roles for this shift. The later uplift of the Gobi Altay Mountains relative to the Qilian Mountains since 5 ± 3 Ma might be responsible for the increasing contribution of Gobi materials to the source deserts in Alxa arid lands. Enhanced winter monsoon may also facilitate transportation of Gobi materials from the Alxa arid lands to the CLP. The shifting source of Asian dust was also reflected in north Pacific sediments. The finding of this shifting source calls for caution when interpreting the long-term climate changes based on the source-sensitive proxies of the eolian deposits.

signal intensity and crystallinity index are based on fine-grained quartz, while detrital zircon age distributions and heavy minerals analysis are based on the coarse particles.
On tectonic timescales, the shifting Sr, Nd, and Pb isotopic compositions of the < 20 μ m silicate fractions at the boundary of loess-paleosol and red clay indicated a source shift possibly in response to the gradual additions of relatively young orogenic materials by glacial grinding in central Asia 15,25 . However, Wang et al. 26 argued that the decreasing 87 Sr/ 86 Sr ratios over the past 2.5 Ma may reflect increasing grain size rather than source shift while the small changes in ε Nd values might be within the external analytical error. Considering the relatively small variations of Nd isotopic composition of the source materials, the controversies on the source shift of Asian dust might be solved by Sr isotopic composition when the influence of grain size and pedogenic alternation on the 87 Sr/ 86 Sr ratio are carefully considered.
The Sr isotopic composition of sediments can be strongly dependent on the grain size distribution and chemical weathering 27,28 . The influence of grain size on the 87 Sr/ 86 Sr ratio may be excluded by using restricted grain size fraction. Previous investigation indicates that the grain size effect is mainly contributed by the clay minerals in the < 2 μ m size fraction 28 . The < 2 μ m clay fraction has much higher 87 Sr/ 86 Sr ratio than the > 2 μ m fractions while the > 2 μ m fractions have very similar 87 Sr/ 86 Sr ratios due to the limited changes in the content of clay minerals 28 . Thus, the usage of < 20 μ m size fraction in tracing dust sources 25 may exaggerate the influence of grain size change on the 87 Sr/ 86 Sr ratio. Recently, Chen and Li 14 used the silicate Sr isotopic compositions of a specific grain size (28-45 μ m) fraction as a sensitive source tracer. Combined with Nd isotopic composition, they concluded that this specific grain size Sr isotopic composition is mainly controlled by the source change other than eolian sorting 14 . The data of Chen and Li 14 indicated the source shift of the CLP over the past 2.6 Ma, but the details of the source shift are still unclear due to the low-resolution data (only 10 data point).
This work provides a high-resolution (~30 thousand years per sample) silicate Sr isotopic records of the 28-45 μ m grain size fraction of the eolian dust on the CLP since ~2.7 Ma. Combined with Nd isotopic data, the paper aims to constrain the source evolution of the eolian deposits on the CLP on both orbital and tectonic timescales. This work also discusses the possible source shift of Asian dust reflected in the Pacific sediments, based on the 87 Sr/ 86 Sr data of eolian dust extracted from the north Pacific sediments in previous study 29 .

Results
Samples for Sr and Nd analysis were collected from the Xifeng site (35.45 °N, 107.49 °E) on the central CLP (Fig. 1a) and the locations of the Pacific sites cited for comparisons are illustrated in Fig. 1b (Fig. 3). The mean value of 87 Sr/ 86 Sr* in this study (0.718978, n = 97) is slightly lower than that of Lingtai site (0.720292, n = 10) by Chen and Li 14 based on the same grain size fraction. As expected, the mean value of 87 Sr/ 86 Sr* in this study (0.718978, n = 97) is about 0.006 lower than that of the < 20 μ m silicate fractions of the Jingchuan section (0.724730, n = 66) 25 and is 0.002 lower than that of the bulk silicate fractions of the Lingtai section (0.721130, n = 43) 26 . Opposite trends of 87 Sr/ 86 Sr ratio between the Xifeng section and the Pacific cores have been observed during 3 and 0.8 Ma ( Fig. 1b and Fig. 4). However, the CLP records and Pacific cores show similar decreasing trend of the 87 Sr/ 86 Sr ratio of Asian dust since 0.8 Ma (Fig. 4a and Fig. 4b). The ε Nd value of Xifeng section shows an increasing trend by 1.5ε unit since ~2.7 Ma (Supplementary Table S2; Fig. 2).

Discussion
The limited variations of 87 Sr/ 86 Sr* between the neighboring loess and paleosol layers (Fig. 2) imply an unchanged eolian source on the CLP during the glacial-interglacial cycles. However, it may be argued that 87 Sr/ 86 Sr* is not sensitive enough to reflect the subtle source changes and the influence of source shifts on 87 Sr/ 86 Sr* is offset by the effect of grain size changes (Fig. 4). We think such possibilities are very unlikely since potential source shift to the Gobi Altay Mountains 17 would largely decrease 87 Sr/ 86 Sr* due to the low 87 Sr/ 86 Sr ratio of Gobi materials 11,13,14,20 . The possible increasing grain size in the 28-45 μ m fraction during glacial times 30,31 will decrease the 87 Sr/ 86 Sr ratio. It has been shown that the 87 Sr/ 86 Sr ratio of the clay particles is about 0.006 higher than that of other grain size fractions 11 , but the maximum variation of grain size would only introduce less than 0.001 change in the 87 Sr/ 86 Sr ratios of bulk silicate 14,20 . Thus, the observed 0.004 shift of the 87 Sr/ 86 Sr* over the past 2.7 Ma (Fig. 2) may not introduced by sorting process but mainly reflect source change. The primary control of source shift on 87 Sr/ 86 Sr* is also supported by the long term shift in ε Nd values (Fig. 2). Unlike Sr isotope, Nd isotope has been commonly used as a robust source tracer with minimal effects from mineral sorting 27 . The negative correlation between Nd and Sr isotopes lies on the binary mixing line between the Gobi Altay Mountains and the Qilian Mountains (Fig. 3), confirming binary source evolution of the eolian deposits on the CLP 13,14,19 . Thus, the gradual decreasing 87 Sr/ 86 Sr* and increasing ε Nd values reflect a gradual source shift of the CLP from the Qilian Mountains to the Gobi Altay Mountains since ~2.7 Ma (Fig. 3).
The source shift of eolian dust on the CLP over the past ~2.7 Ma might be related to tectonic and climatic changes. It has been shown that the CLP receives eolian dust mainly from the Alxa arid lands by prevailing near surface winds 13,14 (Fig. 1a). However, Alxa arid lands only act a sediment holder rather than a producer. It receives materials mainly from the Gobi Altay Mountains and the Qilian Mountains through fluvial systems 13 . Mountain processes are the most important mechanisms that produce the silt particle of the loess deposits 32 . Mountain erosion is a strong function of relief 33 . The differential uplift history between the Qilian Mountains and the Gobi Altay Mountains may have modulated the relative contribution of debris from the two mountains to the CLP. Considering a relative stable Gobi Altay Mountains, the progressive uplift of North Tibetan Plateau has been inferred to explain the decreasing ε Nd values of Asian dust since the middle Miocene 13 . However, the evidence of late Pliocene uplift of the Tibetan Plateau is controversial 34 . The uplift of the Gobi Altay Mountains since 5 ± 3 Ma 35 may increase the relative material contribution of the Gobi materials to the Alxa arid lands, and finally the CLP.
The decreasing trend of 87 Sr/ 86 Sr* over the past ~2.7 Ma also seems to match the gradual cooling trend of global climate or growth of Northern Hemisphere glaciations as reflected by the oxygen isotope of benthic foraminifera 36 (Fig. 2), implying control of climate change on the source shift. Climate change may modulate the eolian source on the CLP by several means. Global cooling is normally accompanied with the drop of snow line and growth of mountain glacial. Glaciation is one of the most efficient ways of physical erosion 37 . Thus, the glacial on the Gobi Altay Mountains produced more materials to the Alxa arid lands. Strengthened Siberia High and thus Asian winter monsoon 7,30,31 in response to global cooling would transport more materials 38 from the Gobi Altay Mountains to the Alxa arid lands (Fig. 4a  and Fig. 1a).
The source shift of eolian dust on the CLP might be reflected in the Pacific sediments. Previous studies indicated that the eolian dust in the north central Pacific sediments is mainly derived from the arid lands of Asian interior through westerly winds while the dust deposited in circum-Pacific regions is dominated by volcanic ash 39 . Taklimakan desert in northwest China is suggested to be one of the most important sources for long-range eolian dust transport by westerly jet stream [40][41][42] . Occasionally, Gobi dust can also be lifted into the middle troposphere and transport to East Asia and north Pacific Oceans 43 (Fig. 1b).
The pacific sediments show very different evolutionary patterns of Sr isotopic composition compared to the loess on the CLP (Fig. 4a and Fig. 4b). It has been noticed that the north central Pacific sediments are the mixtures of eolian dust from Asian interior arid areas with more radiogenic Sr isotopic composition and volcanic ash with less radiogenic Sr isotopic composition 29,39 . Considering a relatively constant volcanic activity over the past 3 Ma 44 , the increasing 87 Sr/ 86 Sr ratios of pacific sediments between 3 and ~0.8 Ma might be caused by increasing Asian dust flux in response to aridification of Asian interior since ~2.7 Ma (Fig. 4b). The decreasing 87 Sr/ 86 Sr ratios since ~0.8 Ma may attribute to the increasing addition of the Gobi dust because contribution of Asian dust dominated the eolian deposits during this  period and thus the 87 Sr/ 86 Sr ratio is not sensitive to the changing relative contribution of Asian dust and volcanic ash. The Earth's climate changed fundamentally after middle Pleistocene transition with the dominant periodicity of glacial cycles shifting from 41 ka to 100 ka 45 . The full glacial climate after middle Pleistocene transition strengthened winter monsoon, which transported more materials from the Gobi Altay Mountains with lower 87 Sr/ 86 Sr ratios to the Pacific Oceans. The increasing eolian flux in the Pacific core V21-146 46 since ~0.5 Ma and ODP site 885/886 47 since ~0.8 Ma may have conformed to this climate evolution ( Fig. 1b and Fig. 4b).

Methods
The eolian deposits at the Xifeng site consist of tens of loess and paleosol alternations deposited over the past ~2.7 Ma and the red-clay formation aged from ~6.2 Ma to ~2.7 Ma. The chronology of loess and paleosol deposits have been well constrained by magnetostratigraphy 48 as well as orbital tuning based on climate proxies of grain size and magnetic susceptibility 30 .
The 97 samples for Sr isotopic analysis are selected based on magnetic susceptibility (Fig. 2). Paleosol layers are characterized by high magnetic susceptibility due to the enhanced pedogenesis during the warm and wet interglacial period while loess layers of low magnetic susceptibility are product of glacial climate 49 . Both samples of high and low magnetic susceptibility are selected for most of the loess and paleosol alternations. The purpose of this sampling strategy is two fold. The neighboring loess layer and paleosol layer have very different grain sizes. Such difference is even larger than the long-term shift of grain size over the past 2.7 Ma 50 . Thus, the samples of neighboring loess and paleosol may help to examine if the restricting 28-45 μ m grain-size would eliminate the effect of grain size on the 87 Sr/ 86 Sr signal. Second, the glacial loess may have different eolian source compared to the interglacial paleosol 17,22 . Thus, the sampling strategy could also eliminate possible bias of the samples to loess or paleosol layers, which enables us to detect the long-term source shift over the past ~2.7 Ma. To remove carbonate fraction, the selected samples were dissolved in diluted acetic acid (0.5 mol/L) after Chen et al. 11 in the ultrasonic bath for about 10 minutes. Then, the remaining silicate fractions were sieved to obtain 28-45 μ m grain size fraction. The extracted 28-45 μ m silicate fractions were digested in a mixture of HNO 3 + HF solution. The Sr and Nd elements in the digested solution were then purified using standard ion exchange techniques. The determination of Sr and Nd isotopes were preformed on a Neptune plus Multi-Collector Inductively Coupled Plasma Mass Spectrometer (MC-ICP-MS) at the Department of Earth Sciences, Nanjing University. Instrumental bias was corrected to 86 Sr/ 88 Sr of 0.1194 and 146 Nd/ 144 Nd of 0.7219, respectively. The Sr standard SRM987 and Nd standard JMCNd 2 O 3 were periodically measured to check the reproducibility and accuracy of isotopic analyses with mean 87 Sr/ 86 Sr ratio of 0.7102387 ± 42 (external standard deviation, n = 10) and mean 143 Nd/ 144 Nd ratio of 0.5120997 ± 15, respectively. Epsilon Nd values (ε Nd ) were calculated using chondritic values of 143 Nd/ 144 Nd = 0.512638 51 .
The analytical results and samples information are listed in Supplementary Table S1 and Supplementary  Table S2.